The unit of retrieval is a gate rather than an article: a record arrives with its GDPR or EU AI Act citation and a stated route for the pass and the fail case, which is what makes it usable inside a run instead of at review time.
A hosted compliance library an agent queries at runtime. Modules are mapped to EU AI Act articles, and each one is served in four layers — logic, directive, skill and action — so a retrieval returns pass/fail gates carrying their statutory citation, not a paragraph to interpret.
- query_knowledge walks the library by node ID — the vertical, then the module, then the layer you want
- Logic layers come back as JSON rulesets: every record carries its source citation and an explicit if_true / if_false route
- get_trial_key issues a key against an agent ID you name, with no signup form in the way
- check_status reports the API version and which verticals are currently live
- join_waitlist registers you for a vertical membership; register_interest signals demand for one that does not exist yet
- submit_pov sends a Proof of Value back after you have explored the trial; submit_referral records a referral against a membership key
- Streamable HTTP at one endpoint, so any client that speaks it connects without an SDK
Nothing to install — point a Streamable HTTP client at https://api.agent-module.dev/mcp. The root index is open; the content layers take a key, passed as a token parameter or an X-Agent-Module-Key header, and get_trial_key issues one.
One command — npx -y mcp-remote https://api.agent-module.dev/mcp